Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
Commit History
-
Install with pnpm; npm deadlocks on the Spaces builder
Three builds died at exactly 8:00 in npm ci. Probes ruled out the obvious causes: the registry answers in under a second, and tarballs pull at 163MB/s, so it is neither reachability nor throughput. npm itself hangs there and the builder kills it at its limit; "Exit handler never called" is what npm prints when SIGTERMed mid-install. pnpm installs the same tree in 10s locally. Lockfile imported from package-lock.json so versions are unchanged, then package-lock.json removed to avoid two lockfiles drifting apart.0ce8cee -
Probe tarball throughput (temporary)
c5358f0 -
Probe the Spaces builder network (temporary)
9f7b515 -
Build the Space on Debian slim
npm ci died with "Exit handler never called" on node:22-alpine after eight minutes on the Spaces builder — an npm bug, not a dependency problem. Debian slim installs the same tree without it.f53c7c1 -
Build the quantization picker
An interactive replacement for the quantization method table in the Transformers docs. The table lists bits; the deciding arithmetic is bytes, and that is where people stall. This does the arithmetic. - lib/data.ts holds the methods and the memory model: weights from params × bits, KV cache from the attention shape, activations as a flat estimate. - lib/hardware.ts is scraped from huggingface.co/hardware, one entry per memory configuration (316 of them) rather than per device, since an RTX 3060 is sold as both 8GB and 12GB. - The model is typed as a Hub repo id and resolved through /api/model, so any model works, not a curated five. - MemoryBar draws the selected precision solid with fp16 behind it as a ghost, so quantizing shows its own effect. - TradeoffMap places the matching methods by when they quantize and whether they serialize — the axes that actually separate them, since every method at a given precision lands at the same footprint. Deploys to a Space via the D…5ae37d4